Organisation Learning Map

Engine: PMLC = python3 <skills-root>/learnings-coach/scripts/pmlc.py (this skill needs the learnings-coach skill installed alongside it). Needs clusters.json (resolve unresolved roles first if time allows; see learning-clusters).

1. Build the learnings (unless a register already exists)

A map of findings alone shows where the evidence sits, not what each level should learn. Build the Learnings Register first by following the learnings-register skill, including tier_messages for every learning. If the user only wants a quick overview, say that the map will show findings only and offer the register as the next step.

2. Build the dashboard

PMLC dashboard <ws> --register <run>/outputs/learnings-register.json \
     [--date <corrected date>] [--title "<plain title>"]      # if the profile flagged summary-sheet errors
#   -> <run>/outputs/organisation-learning-map.html
PMLC interfaces <ws> --top 10

With --register, the HTML has two tabs.

• Overview: the learning × tier matrix (hover a cell for the message; a row links to its detail), a learning plan per level, the control-tier profile of the recommendations, the findings heatmaps (entity × tier, entity × level), and the most frequent factors.
• Detailed register: every register field, each cited finding and evidence item in full, and stakeholder lenses (T1–T5, Investigator, Auditor).

Without --register, it is the single-page Overview. The page is self-contained (offline, light and dark mode) and carries the ProbeMatrix footer. Do not hand-edit the generated HTML; change the register JSON and re-run.

3. Write the learning plan

Write the pack <run>/outputs/organisation-learning-plan.json (schema: learnings-coach/references/pack-schema.md) with these sections:

1. Where the learning sits: three sentences on the shape of the map (which entities carry the most findings, how much sits at supervision and organisational level, which tiers are most involved).
2. What each level must learn: one short section per tier (frontline → board), listing the learnings with that tier's message, priority-1 first.
3. Top learning owners: six to eight clusters with the most priority-1 learnings or highest-rated findings, one sentence each.
4. Strongest interfaces: the top three pairs and what passes between them.
5. Strength of fixes: share of recommendations above the administrative tier, and where stronger fixes are missing.
6. Rollout plan: which audience packs to produce, in which order, and why (usually site leadership and supervisors first, then the largest clusters, then board and auditors). Name the skill for each.

The kinds that fill each numbered section:

1. para.
2. One section per tier, holding learning items: lesson = that tier's message and owner = the cluster, priority-1 first.
3. A table (Cluster, Priority-1 learnings, Why; row_cites).
4. A table (Interface, What passes between them).
5. A para, plus a table of recommendations per control tier.
6. A table (Order, Skill, Audience, Why). A rollout has an order, not dates.

Then:

PMLC check-pack <ws> <run>/outputs/organisation-learning-plan.json   # fix every problem it lists
PMLC pack <ws> <run>/outputs/organisation-learning-plan.json         # -> organisation-learning-plan.html (primary) and .md

Put data-quality corrections in data_quality, findings left out on purpose in exclusions, and same-condition findings in supporting_cites. Keep the plan to three or four pages. The dashboard and register carry the detail.